| 1 | /* |
| 2 | * librpma_fio: librpma_apm and librpma_gpspm engines' common header. |
| 3 | * |
| 4 | * Copyright 2021, Intel Corporation |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU General Public License, |
| 8 | * version 2 as published by the Free Software Foundation.. |
| 9 | * |
| 10 | * This program is distributed in the hope that it will be useful, |
| 11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 13 | * GNU General Public License for more details. |
| 14 | */ |
| 15 | |
| 16 | #ifndef LIBRPMA_FIO_H |
| 17 | #define LIBRPMA_FIO_H 1 |
| 18 | |
| 19 | #include "../fio.h" |
| 20 | #include "../optgroup.h" |
| 21 | |
| 22 | #include <librpma.h> |
| 23 | |
| 24 | /* servers' and clients' common */ |
| 25 | |
| 26 | #define librpma_td_verror(td, err, func) \ |
| 27 | td_vmsg((td), (err), rpma_err_2str(err), (func)) |
| 28 | |
| 29 | /* ceil(a / b) = (a + b - 1) / b */ |
| 30 | #define LIBRPMA_FIO_CEIL(a, b) (((a) + (b) - 1) / (b)) |
| 31 | |
| 32 | /* common option structure for server and client */ |
| 33 | struct librpma_fio_options_values { |
| 34 | /* |
| 35 | * FIO considers .off1 == 0 absent so the first meaningful field has to |
| 36 | * have padding ahead of it. |
| 37 | */ |
| 38 | void *pad; |
| 39 | char *server_ip; |
| 40 | /* base server listening port */ |
| 41 | char *port; |
| 42 | /* Direct Write to PMem is possible */ |
| 43 | unsigned int direct_write_to_pmem; |
| 44 | /* Set to 0 to wait for completion instead of busy-wait polling completion. */ |
| 45 | unsigned int busy_wait_polling; |
| 46 | }; |
| 47 | |
| 48 | extern struct fio_option librpma_fio_options[]; |
| 49 | |
| 50 | /* |
| 51 | * Limited by the maximum length of the private data |
| 52 | * for rdma_connect() in case of RDMA_PS_TCP (28 bytes). |
| 53 | */ |
| 54 | #define LIBRPMA_FIO_DESCRIPTOR_MAX_SIZE 24 |
| 55 | |
| 56 | struct librpma_fio_workspace { |
| 57 | uint16_t max_msg_num; /* # of RQ slots */ |
| 58 | uint8_t direct_write_to_pmem; /* Direct Write to PMem is possible */ |
| 59 | uint8_t mr_desc_size; /* size of mr_desc in descriptor[] */ |
| 60 | /* buffer containing mr_desc */ |
| 61 | char descriptor[LIBRPMA_FIO_DESCRIPTOR_MAX_SIZE]; |
| 62 | }; |
| 63 | |
| 64 | #define LIBRPMA_FIO_PORT_STR_LEN_MAX 12 |
| 65 | |
| 66 | int librpma_fio_td_port(const char *port_base_str, struct thread_data *td, |
| 67 | char *port_out); |
| 68 | |
| 69 | struct librpma_fio_mem { |
| 70 | /* memory buffer */ |
| 71 | char *mem_ptr; |
| 72 | |
| 73 | /* size of the mapped persistent memory */ |
| 74 | size_t size_mmap; |
| 75 | }; |
| 76 | |
| 77 | char *librpma_fio_allocate_dram(struct thread_data *td, size_t size, |
| 78 | struct librpma_fio_mem *mem); |
| 79 | |
| 80 | char *librpma_fio_allocate_pmem(struct thread_data *td, const char *filename, |
| 81 | size_t size, struct librpma_fio_mem *mem); |
| 82 | |
| 83 | void librpma_fio_free(struct librpma_fio_mem *mem); |
| 84 | |
| 85 | /* clients' common */ |
| 86 | |
| 87 | typedef int (*librpma_fio_flush_t)(struct thread_data *td, |
| 88 | struct io_u *first_io_u, struct io_u *last_io_u, |
| 89 | unsigned long long int len); |
| 90 | |
| 91 | /* |
| 92 | * RETURN VALUE |
| 93 | * - ( 1) - on success |
| 94 | * - ( 0) - skip |
| 95 | * - (-1) - on error |
| 96 | */ |
| 97 | typedef int (*librpma_fio_get_io_u_index_t)(struct rpma_completion *cmpl, |
| 98 | unsigned int *io_u_index); |
| 99 | |
| 100 | struct librpma_fio_client_data { |
| 101 | struct rpma_peer *peer; |
| 102 | struct rpma_conn *conn; |
| 103 | |
| 104 | /* aligned td->orig_buffer */ |
| 105 | char *orig_buffer_aligned; |
| 106 | |
| 107 | /* ious's base address memory registration (cd->orig_buffer_aligned) */ |
| 108 | struct rpma_mr_local *orig_mr; |
| 109 | |
| 110 | struct librpma_fio_workspace *ws; |
| 111 | |
| 112 | /* a server's memory representation */ |
| 113 | struct rpma_mr_remote *server_mr; |
| 114 | enum rpma_flush_type server_mr_flush_type; |
| 115 | |
| 116 | /* remote workspace description */ |
| 117 | size_t ws_size; |
| 118 | |
| 119 | /* in-memory queues */ |
| 120 | struct io_u **io_us_queued; |
| 121 | int io_u_queued_nr; |
| 122 | struct io_u **io_us_flight; |
| 123 | int io_u_flight_nr; |
| 124 | struct io_u **io_us_completed; |
| 125 | int io_u_completed_nr; |
| 126 | |
| 127 | /* SQ control. Note: all of them have to be kept in sync. */ |
| 128 | uint32_t op_send_posted; |
| 129 | uint32_t op_send_completed; |
| 130 | uint32_t op_recv_completed; |
| 131 | |
| 132 | librpma_fio_flush_t flush; |
| 133 | librpma_fio_get_io_u_index_t get_io_u_index; |
| 134 | |
| 135 | /* engine-specific client data */ |
| 136 | void *client_data; |
| 137 | }; |
| 138 | |
| 139 | int librpma_fio_client_init(struct thread_data *td, |
| 140 | struct rpma_conn_cfg *cfg); |
| 141 | void librpma_fio_client_cleanup(struct thread_data *td); |
| 142 | |
| 143 | int librpma_fio_file_nop(struct thread_data *td, struct fio_file *f); |
| 144 | int librpma_fio_client_get_file_size(struct thread_data *td, |
| 145 | struct fio_file *f); |
| 146 | |
| 147 | int librpma_fio_client_post_init(struct thread_data *td); |
| 148 | |
| 149 | enum fio_q_status librpma_fio_client_queue(struct thread_data *td, |
| 150 | struct io_u *io_u); |
| 151 | |
| 152 | int librpma_fio_client_commit(struct thread_data *td); |
| 153 | |
| 154 | int librpma_fio_client_getevents(struct thread_data *td, unsigned int min, |
| 155 | unsigned int max, const struct timespec *t); |
| 156 | |
| 157 | struct io_u *librpma_fio_client_event(struct thread_data *td, int event); |
| 158 | |
| 159 | char *librpma_fio_client_errdetails(struct io_u *io_u); |
| 160 | |
| 161 | static inline int librpma_fio_client_io_read(struct thread_data *td, |
| 162 | struct io_u *io_u, int flags) |
| 163 | { |
| 164 | struct librpma_fio_client_data *ccd = td->io_ops_data; |
| 165 | size_t dst_offset = (char *)(io_u->xfer_buf) - ccd->orig_buffer_aligned; |
| 166 | size_t src_offset = io_u->offset; |
| 167 | int ret; |
| 168 | |
| 169 | if ((ret = rpma_read(ccd->conn, ccd->orig_mr, dst_offset, |
| 170 | ccd->server_mr, src_offset, io_u->xfer_buflen, |
| 171 | flags, (void *)(uintptr_t)io_u->index))) { |
| 172 | librpma_td_verror(td, ret, "rpma_read"); |
| 173 | return -1; |
| 174 | } |
| 175 | |
| 176 | return 0; |
| 177 | } |
| 178 | |
| 179 | static inline int librpma_fio_client_io_write(struct thread_data *td, |
| 180 | struct io_u *io_u) |
| 181 | { |
| 182 | struct librpma_fio_client_data *ccd = td->io_ops_data; |
| 183 | size_t src_offset = (char *)(io_u->xfer_buf) - ccd->orig_buffer_aligned; |
| 184 | size_t dst_offset = io_u->offset; |
| 185 | int ret; |
| 186 | |
| 187 | if ((ret = rpma_write(ccd->conn, ccd->server_mr, dst_offset, |
| 188 | ccd->orig_mr, src_offset, io_u->xfer_buflen, |
| 189 | RPMA_F_COMPLETION_ON_ERROR, |
| 190 | (void *)(uintptr_t)io_u->index))) { |
| 191 | librpma_td_verror(td, ret, "rpma_write"); |
| 192 | return -1; |
| 193 | } |
| 194 | |
| 195 | return 0; |
| 196 | } |
| 197 | |
| 198 | static inline int librpma_fio_client_io_complete_all_sends( |
| 199 | struct thread_data *td) |
| 200 | { |
| 201 | struct librpma_fio_client_data *ccd = td->io_ops_data; |
| 202 | struct rpma_completion cmpl; |
| 203 | int ret; |
| 204 | |
| 205 | while (ccd->op_send_posted != ccd->op_send_completed) { |
| 206 | /* get a completion */ |
| 207 | ret = rpma_conn_completion_get(ccd->conn, &cmpl); |
| 208 | if (ret == RPMA_E_NO_COMPLETION) { |
| 209 | /* lack of completion is not an error */ |
| 210 | continue; |
| 211 | } else if (ret != 0) { |
| 212 | /* an error occurred */ |
| 213 | librpma_td_verror(td, ret, "rpma_conn_completion_get"); |
| 214 | break; |
| 215 | } |
| 216 | |
| 217 | if (cmpl.op_status != IBV_WC_SUCCESS) |
| 218 | return -1; |
| 219 | |
| 220 | if (cmpl.op == RPMA_OP_SEND) |
| 221 | ++ccd->op_send_completed; |
| 222 | else { |
| 223 | log_err( |
| 224 | "A completion other than RPMA_OP_SEND got during cleaning up the CQ from SENDs\n"); |
| 225 | return -1; |
| 226 | } |
| 227 | } |
| 228 | |
| 229 | /* |
| 230 | * All posted SENDs are completed and RECVs for them (responses) are |
| 231 | * completed. This is the initial situation so the counters are reset. |
| 232 | */ |
| 233 | if (ccd->op_send_posted == ccd->op_send_completed && |
| 234 | ccd->op_send_completed == ccd->op_recv_completed) { |
| 235 | ccd->op_send_posted = 0; |
| 236 | ccd->op_send_completed = 0; |
| 237 | ccd->op_recv_completed = 0; |
| 238 | } |
| 239 | |
| 240 | return 0; |
| 241 | } |
| 242 | |
| 243 | /* servers' common */ |
| 244 | |
| 245 | typedef int (*librpma_fio_prepare_connection_t)( |
| 246 | struct thread_data *td, |
| 247 | struct rpma_conn_req *conn_req); |
| 248 | |
| 249 | struct librpma_fio_server_data { |
| 250 | struct rpma_peer *peer; |
| 251 | |
| 252 | /* resources of an incoming connection */ |
| 253 | struct rpma_conn *conn; |
| 254 | |
| 255 | char *ws_ptr; |
| 256 | struct rpma_mr_local *ws_mr; |
| 257 | struct librpma_fio_mem mem; |
| 258 | |
| 259 | /* engine-specific server data */ |
| 260 | void *server_data; |
| 261 | |
| 262 | librpma_fio_prepare_connection_t prepare_connection; |
| 263 | }; |
| 264 | |
| 265 | int librpma_fio_server_init(struct thread_data *td); |
| 266 | |
| 267 | void librpma_fio_server_cleanup(struct thread_data *td); |
| 268 | |
| 269 | int librpma_fio_server_open_file(struct thread_data *td, |
| 270 | struct fio_file *f, struct rpma_conn_cfg *cfg); |
| 271 | |
| 272 | int librpma_fio_server_close_file(struct thread_data *td, |
| 273 | struct fio_file *f); |
| 274 | |
| 275 | #endif /* LIBRPMA_FIO_H */ |